Yes, a CT scan can cause cancer, but the risk from any one medically justified scan is very small and is usually far smaller than the danger of missing what the scan is looking for. That two-part answer is the whole story: the radiation is real, and so is the benefit, and honest decision-making means holding both at once rather than fixating on either alone.
The reason this question feels alarming right now is a single number. In 2025, headlines reported that CT imaging in the United States could eventually drive about 103,000 cancers. Understanding what that figure is, and just as importantly what it is not, is the difference between useful caution and needless dread.
Key points#
- A projected 103,000 future cancers is a statistical model output, not a tally of diagnosed tumors.
- The number comes from applying the linear no threshold (LNT) assumption to roughly 93 million scans.
- LNT is a sensible safety-planning rule, but using it to compute a precise national body count is contested.
- The best measured evidence (the EPI-CT study) confirms real risk in children scanned young, not a specific adult headcount.
- For one patient on one day, the question that matters is whether the scan will change what happens next.
What kind of number is 103,000?#
The estimate comes from a study led by Rebecca Smith-Bindman and colleagues, published in JAMA Internal Medicine in April 2025 and summarized by NIH Research Matters. The researchers took the roughly 93 million CT examinations done on about 61.5 million patients in 2023 and estimated that they could eventually produce around 103,000 radiation-related cancers, with a 90 percent uncertainty range of roughly 96,400 to 109,500. Projected forward, that would be close to 5 percent of new cancer diagnoses each year, most often lung, colon, and leukemia. Per-scan risk ran higher in children, but because adults are imaged far more often, adults made up the bulk of the projected total.
Here is how that figure was built. It came out of the National Cancer Institute's Radiation Risk Assessment Tool (RadRAT), which relies on risk models from the National Academies' BEIR VII report. In other words, 103,000 is not something anyone counted. It is an extrapolation: an estimate of what would happen if a specific dose-to-risk relationship stayed perfectly straight all the way down to the tiny doses of a single scan.
The assumption doing the heavy lifting#
That straight line is the linear no threshold model. LNT holds that cancer risk climbs in direct proportion to radiation dose, with no floor below which the risk drops to zero. Its firmest evidence comes from groups that received relatively high doses, most notably survivors of the atomic bombings in Japan, where excess cancers are plainly measurable. LNT takes that well-measured, high-dose relationship and extends the line downward into the low-dose region, where direct measurement becomes very hard.
Think of it like a dimmer switch you can only test at the bright end. You can see clearly how brightness tracks the dial from half to full, and LNT assumes the same proportional behavior continues all the way down to a faint glow, even though the glow is too dim to read directly.
Radiation-protection bodies such as the International Commission on Radiological Protection (ICRP) adopt LNT as a prudent default for setting dose limits. When the goal is to keep occupational and public radiation as low as reasonably achievable, a no-safe-dose assumption is exactly the cautious choice you want. The disagreement is not about whether LNT is a good planning rule. It is about whether the same rule should be run in reverse to turn millions of tiny individual risks into one confident national number.
Why a modeled estimate is not measured harm#
That objection is the argument of a 2025 viewpoint in the American Journal of Respiratory and Critical Care Medicine, whose title states outright that the 5 percent figure is overstated. The authors call the projection largely a modeling artifact: a theoretical ceiling produced by pushing older risk models well past the dose range where their data are solid.
Their central methodological complaint is about collective dose, the practice of adding up minuscule per-person risks across an enormous population to manufacture a large sum. ICRP itself warns against using collective dose this way to compute a number of deaths, precisely because the individual risks being added together sit below the level at which epidemiology can actually confirm them. Below roughly 100 millisievert, the excess cancer signal is faint and easily drowned out by the background rate of ordinary cancers and by statistical noise.
None of this means CT radiation is harmless. It means a projection inherits the uncertainty of its assumptions, and a number written to three significant digits can suggest a precision the science does not yet support.
What the real-world data actually show#
Between the two positions sits the strongest measured evidence. The EPI-CT study, coordinated by the International Agency for Research on Cancer, followed roughly 948,000 people across nine European countries who were scanned before age 22. Published in 2023, it found a genuine dose-response: the risk of blood cancers rose with cumulative bone-marrow dose, with an excess relative risk near 1.96 per 100 milligray, close to a tripling of risk at that level. That is observed harm, not a forecast, and it argues against the idea of a clean safe threshold, at least in children, whose dividing tissues are more radiosensitive.
Read carefully, though, EPI-CT describes cumulative dose in a young, sensitive population tracked over years. It supports the direction of the LNT assumption without validating any single national headcount for adults, most of whom receive one scan at a time for one specific clinical question.
Weighing it for a real decision#
The trap is to apply a population statistic to a personal choice. They answer different questions.
On the day a scan is ordered, the useful question is simple: will the images change what happens next? A CT that confirms or rules out a stroke, a blood clot, or a tumor delivers a benefit that dwarfs a per-scan cancer risk usually estimated at a small fraction of one percent. The population figure becomes worrying mainly in the aggregate, when scans are duplicated, ordered by reflex, or done where a lower-dose test would have answered the same question just as well.
Only two levers genuinely lower risk. The first is justification: the scan should be clinically warranted in the first place. The second is optimization: modern equipment and protocols should keep the dose as low as the diagnostic task allows. Neither lever asks a frightened patient to refuse a scan they actually need.
Sources and further reading
Questions and answers
Should I refuse a CT scan because of the cancer risk?
Not on the basis of the headline number alone. For a scan your clinician judges necessary, the information it provides almost always outweighs a very small radiation risk. A reasonable question to ask instead is whether the scan will change your management, or whether a lower-dose alternative such as ultrasound or MRI would answer the same question.
Does the 103,000 figure mean 103,000 people will definitely get cancer from CT?
No. It is a modeled projection built on the linear no threshold assumption, carried down to doses where the effect cannot be measured directly. It is best read as an argument against unnecessary imaging across a whole health system, not as a diagnosis for any individual.
Are children more vulnerable to CT radiation than adults?
Yes. Children's dividing tissues are more radiosensitive, and the EPI-CT study measured a real rise in blood-cancer risk with cumulative dose in people scanned young. This is a major reason pediatric imaging protocols are deliberately dose-reduced. Weighed honestly, the 103,000 figure is most useful as a case against scans that were never needed. It should sharpen judgment about which images to order while leaving room to accept a scan that genuinely helps.